memory, and in long term memory. We are not conscious of these processes. What we

are conscious of, to a high degree, is language form on one hand, and experiencing

'meaning' on the other. The effect is magical; as soon as we have form, we also have

meaning, with no awareness of the intervening cognition. Introspectively, our experience

in this regard is analogous to perception - we see an object because it is there, we

understand a sentence instantly because it 'has' that meaning. This remarkable and

invisible efficiency of our meaning assigning capacities drives our folk theories about

language, which conflate form and meaning, just as folk theories about the world conflate

existence and perception.

The technical description of mental space phenomena is developed in a number of

publications.1 I would like to attempt, in this article, to give a more informal view of the

phenomena, their complexity, and their importance for psychologists interested in language

and thought.

AN EXAMPLE - CONNECTORS AND ACCESS PATHS

Consider the simple statement Max thought the winner received $100. Perhaps,

the most obvious way to understand this is to assume that there was a contest, that prizes

were given out, that one person won the contest and received a prize, and that Max, who

was aware of all this, believed that prize to have been $100. But the statement by itself

contains none of this. It just fits a plausible scenario, which our background knowledge

makes available. We shall see that it also fits many other scenarios, some of which might

1 Fauconnier (1994, 1997), Cutrer (1994), Fauconnier and Sweetser (1996), Fauconnier and Turner(in press).

be more appropriate in some contexts, and some of which are extremely implausible. What

exactly then is the sentence telling us?

An important part of what the language form is doing is to prompt us to set up

mental spaces, elements, and connections between them. The form provides key

indications about the mental process we need to engage in, but it is also deeply

underspecified, which accounts for the multiplicity of scenarios that it may successfully fit.

An interesting challenge for cognitive science at large is to understand how this

underespecification can be resolved, how and why in certain situations certain mental space

configurations are unconsciously chosen over others, and how the understander converges

on specific appropriate scenarios and connection patterns.

Mental spaces are small conceptual packets constructed as we think and talk, for

purposes of local understanding and action. They are very partial assemblies containing

elements, and structured by frames and cognitive models. They are interconnected and can

be modified as thought and discourse unfold.

In our example, two mental spaces are set up.2 One is the Base space, B, the inital

space with partial structure corresponding to what has already been introduced at that point

in the discourse, or what may be introduced freely because it is pragmatically available in

the situation. Another mental space, M, subordinate to this one will contain partial

structure corresponding to 'what Max thinks'. It is structured by the form '_received

$100' (the subordinate compement clause of thought.) That form evokes a general frame

<x receive y>, of which we may know a great number of more specific instances (receive

money, a shock, a letter, guests, ...). The expression Max thought is called a space-

builder, because it explicitly sets up the second mental space. Max and the winner are noun

2 I simplify the account here for expository purposes, by leaving out additional spaces set up bythe past tense operators.

phrases and will provide access to elements in the spaces. This happens as follows: the

noun phrase is a name or description which either fits some already established element in

some space, or introduces a new element in some space. That element in turn may provide

access to another element, through a cognitive connection (called a connector), as we shall

illustrate and explain. Elements are conceived of as high order mental entities (like nodes in

Shastri's models3). They may or may not themselves refer to objects in the world. In our

example, the name Max accesses an element a in B, the base space (which intuitively might

correspond to a real or fictive person called Max). The description the winner accesses an

element w that we call a role, and is assumed to belong to a general frame of winning ( w

wins ), and also to a more specific instance of that frame appropriate for the given context

(winning a particular race, lottery, game, etc.). Roles can have values, and a role element

can always access another element which is the value of that role. So we can say The

winner will get $10, without pointing to any particular individual. This is the role

interpretation. Or we can sayThe winner is bald, where being bald is a property of the

individual who happened to win, not a condition for getting the prize. This is the value

interpretation. Roles, then, are linked cognitively to their values by a role-value connector.

The two mental spaces B and M are connected. There can be counterparts of

elements of B in M. For example, if we said Max thinks he will win, intending the

pronoun he to refer back to Max, then space B would contain a (for Max) and space M

('Max thinks__') would contain a counterpart a' of a. In space M, a relation equivalent to

'a' wins' would be satisfied, while the same relation would not necessarily be satisfied in

the base B for a (the counterpart of a'). It is convenient to use mental space diagrams of

the following form to represent this evolving structure:

3 Shastri and Grannes (1996).

a

a'

Base space B

Belief space M

a' WIN

a: NAME Max

An important principle defines a general procedure for accessing

elements:

Access Principle

If two elements a and a' are linked by a connector F ( a' = F(a) ), then element

a' can be identified by naming, describing, or pointing to, its counterpart a.

This very general principle applies to all types of connectors across and within

mental spaces (identity, analogy, metaphor, metonymy, role/value). A simple example

of its application would be Max thinks Harry's name is Joe. An element e associated

with the name Harry is set up in the base. Its counterpart e' in M satisfies <e' named

Joe>. And e' is accessed by means of its counterpart e in the base, using the name

Harry associated with e. In other words, even though Harry is the appropriate name in

one space, it can be used to access the corresponding element in another space, where

another name is appropriate.

Returning now to the original example, let's see what the accessing possibilities

might be. Suppose first that the Access principle does not apply at all. Then the

description the winner must identify a role w directly in space M. The structure added to M

by the sentence is <w received $100>. This is a pure role interpretation within space M. It

does not access any corresponding value for that role. Therefore the interpretation is that

Max thought there was a contest, and that he thought it was a feature of that contest to

award $100 to whoever wins. This accessing strategy is non-committal as to whether the

speaker also assumes there was such a contest, and as to whether an actual winner was

ever selected, or as to whether Max thinks that a winner was selected. The sentence under

this strategy would be appropriate in a variety of contexts. For example, mini-discourses

like the following could include the above strategy:

The Boston marathon will take place next week. Max thought the winner received

$100, but it turns out there won't be any prize money.

My friends were under the impression that I was running a lottery in my garage.

Max thought the winner received $100. But they were all wrong, there was no lottery.

Suppose now that the Accessing principle operates, but only within the subordinate

space M, linking the role 'winner' to a value of that role, b. As before, Max believes that

there was a contest, and moreover that somebody won and he has additional beliefs about

the person who he assumes won. The structure in M is now <b receive $100>. Although

a likely default is that the $100 was prize money, this is no longer imposed. Max may

believe that something else happened, causing the person who had won to receive $100

independently. In plain English, this is illustrated most distinctly by a situation in which

(according to the speaker) Max believes (perhaps incorrectly) that Susan won a race, and

that she also independently got $100 for the used car that she was selling.

In the two accessing possibilities just considered, a word (winner) simply evokes a

script within a single space. But the Access principle may also operate across spaces. The

speaker may have a particular contest in mind, for which there is a role winner , set up as

an element w in the base B. That role can have a value a (for example Harry) with a

counterpart a' in M. The Access principle allows the winner to access w, and then its

value a, and finally the counterpart of a, element a' in the subordinate space M. The

structure set up in M is <a' receive $100>. The interpretation is that the speaker presents

Harry as the winner, and says that Max thought Harry received $100. This is compatible

with Max knowing nothing about the contest, or believing that someone other than Harry

won, and that Harry got the $100 for selling the used car, or as a consolation prize in the

contest.

Other accessing paths are available for this simple sentence, if counterpart roles are

considered, or if the extra space introduced by the past tense is taken into account, or if

other spaces are accessible at that point in the discourse to provide counterparts. Typically

an understander does not have to consider all possibilities. The intended path will be

favored by the space configuration in the discourse at the point when the statement is made.

We might already have the role 'winner' in space B, or in space M but not B. We might

already have different values in M and B for the 'same' role (w and its counterpart w'),

and so on. In other cases, of course, the understander will lack sufficient information, and

may have to revise a space configuration, or may simply misunderstand a speaker's intent.

Our example had a subordinate space corresponding to 'belief'. There are many

other kinds of spaces, but they all share these complex accessing possibilities. For

instance, time expressions are space-builders and set up new spaces in discourse.

Consider (as part of a larger discourse) In 1968, the winner received $100. As before, we

have a base B and a subordinate space M, corresponding to 1968, set up by the space-

builder in 1968. And as before, the noun phrase the winner can access a role in M, or the

counterpart role in M of a role in B, or the counterpart in M of a value in B, or a value of a

role in M. Situations that fit these respective strategies might be:

- there was a certain type of game in 1968 (no longer played today) in which you got $100

for winning;

- there is a certain sports competition, say the Boston marathon, which exists today (role

w in space B) and also existed in 1968 (counterpart role w' in space M). In 1968 (as

opposed to today), whoever won got $100;

- the winner of the chess championship held today is Susan; back in 1968, in unrelated

circumstances (e.g. selling her used car), Susan received $100;

- the winner back in 1968 of the contest we are talking about was Harry, and that year,

Harry received $100 (perhaps for selling his used car).

The access paths available in this example involving time (1968) are exactly the

same as the access paths in the previous example involving belief. Even though time and

belief are conceptually quite different, they give rise, at the level of discourse management

considered here, to the same mental space configurations. More generally, we find that

mental spaces are set up for a wide variety of conceptual domains that include time, belief,

wishes, plays, movies, pictures, possibility, necessity, hypotheticals and counterfactuals,

locatives, and reality. The connectors, the access principle, the role/value distinctions work

uniformly across this broad range of cases.

The choice of an access path and of particular connectors is underspecified (or

sometimes not specified at all) by the language forms. This is not vagueness, however. A

speaker has a particular path in mind, and it must be recovered by the understander. Paths

can be forced by elaboration. For instance, in our example case above, one might have

said:

In 1968, the winner received $100, but Harry, although he won, only got $50 ,

because he was fined for yelling at the umpire.

This will force the role reading inside space M: the rules of the game at the time

were that the winner got $100. The rest of the statement is about the particular individual

who won, and the fact that *he* only got $50. This could also have been expressed using

a role/value link: That year, the winner only got $50, because he was fined for yelling at the

umpire.

Such examples illustrate (very partially) the kinds of accessing strategies that may

be available, and that may have to be reconstructed in understanding what is meant. An

important feature of such constructions from a psychological and processing point of view

is that in many cases, the access paths will collapse,because counterparts in the relevant

spaces do not differ from each other. This is in fact the default assumption that we

typically make in the absence of explicit information to the contrary. We call this default

principle 'optimization' of spaces. So, for example, using our winner example once more,

Max thought the winner received $100, if spaces B and M are identically structured in

terms of background assumptions and connections (existence and nature of the contest,

identity of the winner, focus on the nature of the prize), there will no longer be any

difference between the interpretations provided by the connecting paths. The role 'winner'

and its value (say Harry) will be 'the same' in both spaces, and the import of the statement,

with any connecting path, will be the same. It will be about the prize amount that Harry,

qua winner, received in the contest. Notice that this is in fact the interpretation that first

came to mind when we saw the sentence presented as a statement in isolation. Since there

had been no prior discourse structure set up, we directly made the most simple (i.e.

optimized) default assumptions, and obtained a sensible interpretation. The hidden

assumption we made was that the two spaces (base and belief) had matching structure. In

addition, we completed the resulting configuration by means of the most obvious available

background scenario (contest with prize money) that would fit 'winner' and 'receive $100'.

However, as soon as a sentence is part of extended discourse, it will be prompting

strategies within the mental space configuration already set up at the time when that

sentence comes into the discourse. Explicit links and structures present at that time in the

0

meaning-building process will constrain the construction of accessing paths and the

imposition of scenarios. In particular, such existing discourse connections may prevent

optimization, default strategies, or the adoption of default scenarios.

Crucially, then, the apparently 'simple' or 'prototypical' cases are only special

(default) instances of the general space-building operations. The drawback of studying

sentences in isolation (as a linguist or as a psycholinguist) is that only the defaults will

emerge. Those defaults, far from helping us understand the general strategies, actually

occlude them from the observer by effectively conflating them.

MODALITY - THE CASE OF ASL

Spoken languages offer considerable evidence for mental space organization. But

interestingly, independent evidence is also available from sign languages such as ASL,

which operate in a different modality, visual-gestural rather than oral-auditory. Van Hoek

(1996), Liddell (1995a,b), Poulin (1996) are among those who have very successfully

pursued an approach initiated by Richard Lacy in unpublished work in the late seventies.

Their research has provided extensive evidence for mental space constructions in ASL. As

Liddell demonstrates, sign languages additionally make use of grounded mental spaces in

their grammars, by taking advantage of the spatial modality.

The clearest example of this is the signing space set up by signers in order to

perform various referential and conceptual operations. As Scott Liddell writes: "Sign

languages are well known for their ability to create, as part of the most ordinary discourse,

elaborate conceptual representations in the space in front of the signer. Because of the

importance of space in ordinary signed discourse, signed languages have come to be

structured in ways which take advantage of those spatial representations. Pronouns and

some types of verbs can be produced at specific locations in space or directed towards

1

specific areas of space to produce distinctive meanings. Signs of this type can also be

directed toward things that are physically present, including the signer, the addressee, other

participants, and other entities. ... The linguistic uniqueness of the ability to make semantic

distinctions by producing signs toward an apparently unlimited number of locations is

beyond question." [Liddell (1995b)].

The physical signing space with referential loci that one can point to serves to

ground a corresponding mental space in which elements are being introduced and

structured. Subspaces can then be set up with overt counterpart structure analogous to the

mental space connections described above for our English example. Strikingly, the Access

principle operates transparently in such cases. As Karen Van Hoek shows, one can point

to loci in order to access the counterparts in some space of the elements corresponding to

those loci. The choice of accessing strategies is particularly interesting, since it depends on

subtle distinctions having to do with focus, viewpoint, and the ultimate goals of the

conversational exchange. Here is an example from Van Hoek's work. The notation is

explained below the transcription.

FIG 1

In this short piece of discourse, two different loci are set up in the signing space for

the same individual (the beautiful friend). One locus is to the left of the signer -

and corresponds to present time and to the Paris location. The other locus, to the right of

the speaker -

2

corresponds to past time, with location in Texas. In almost all of the discourse, the

present/Paris locus is used to access "the friend", except in the part corresponding to "I

knew that she would be ugly all her life", where the point of view shifts to the past, and the

other locus (past/Texas) is used. Then it shifts back to present time, and the present locus

("I see, she's beautiful, ..). Immediately after this passage, the signer was asked more

about his memories and went on with the following discourse, which shifted viewpoint and

focus back to the past, and now accessed the same individual (the beautiful friend) from the

past locus (to the right of the signer) .

FIG 2

With examples like these and many others, Van Hoek shows that, just like in our

'1968' example, the elements in one mental space may be accessed from the referential

locus in the signing space appropriate for that particular mental space (e.g. past), or from a

locus for its counterpart in some higher space (e.g. present/Base). The spatial modality

allows the spaces to be grounded: one can actually point or direct other signs toward one

or the other referential locus, as one would in pointing deictically at relevant objects,

physically present in the context. Liddell shows how the manipulation of such grounded

spaces (token space, surrogate space, and real space) is incorporated into the grammar of

ASL to yield intricate reference mechanisms. Poulin (1996) shows how such spaces can

be shifted to reflect changes in viewpoint or epistemic stance. This is typically

accomplished physically by body shifts, and repositioning:

3

FIG 3

In A, the two referents i and j are set up as loci in the signing space in front of the signer.

In B, there is a body shift to the right, and the signer identifies with referent i (taking the

viewpoint of i so to speak). In C, there is a body shift to the left, and the signer now

identifies with the other referent, j.

Liddell (1995b) shows in great detail the link between such referential processes

incorporated into ASL grammar, and general linguistic and non-linguistic mental space

building and grounding.

The relevant language universals here are the modality-independent principles of

connections and access across mental spaces. The modality-specific universals are the

ways in which these mental configurations can be indicated through language (spoken or

signed). In both spoken and signed languages, we find grammatical devices for building

spaces (adverbials, subject-verb combinations, conjunctions, ...); in spoken language,

pronominal systems and other anaphoric devices code linearly the construction or

reactivation of mental space elements. In sign language, the same effect is achieved by

constructing grounded spaces, which take advantage of the spatial modality.

TENSE AND MOOD

The spaces we have been talking about are set up dynamically throughout an

ongoing discourse, on the basis of linguistic and non-linguistic clues and information. The

general scheme is one of new spaces built relative to existing ones:

4B (base space)

M

M M

M

M M

M

M M

M

12 i

11 1221 22

il ij

ijk

. . .

. . .

A piece of discourse will start with a base B. Space M1 is then set up subordinate

to B, then space M11, subordinate to M1, and so on. Returning to the base B, one can open

space M2, than M21, etc., return to B a number of times, opening spaces Mi, and daughter

spaces Mij, Mijk, and so on.

At any given stage of the discourse, one of the spaces is a base for the system, and

one of the spaces (possibly the same one) is in focus. Construction at the next stage will

be relative either to the Base Space or to the Focus Space.4 The discourse moves through

4 This is the scheme developed in Dinsmore (1991).

5

the lattice of spaces; viewpoint and focus shift as we go from one space to the next. But at

any point, the Base Space remains accessible as a possible starting point for another

construction.

Dinsmore (1991) and Cutrer (1994) have shown that a major function of tense in

language is to establish local time ordering relations between neighboring mental spaces,

and to keep track of viewpoint and focus shifts. Cutrer (1994) develops a sophisticated set

of principles for mental space connections guided by tense, and explains thereby many

mysterious features of the ways in which we construct time and viewpoint organization

with language. We cannot, here, go into the mechanics of tense and time, but the

following example, borrowed from Fauconnier (1997) will help to give an informal idea of

what is going on.

The example is a very short piece of discourse:

Max is 23. He has lived abroad. In 1990, he lived in Rome. In 1991

he would move to Venice. He would then have lived a year in Rome.

The space building dynamics associated with the production and/or

understanding of this mini-story run as follows:

1) We start with a single space, which is the Base, and also the initial

Viewpoint and Focus. We structure that space with the information that Max is

23 years old.

2) Keeping that space in Focus, we add the (present) information that

Max has lived abroad. This information is presented via a past Event space

('Max live abroad').

3) In the next sentence, in 1990 is a space builder. It sets up a new

Focus space, in which we build the content 'Max live in Rome'. This is also

6

the new Event space, since we are considering the event/state of Max living in

Rome.

4) This Focus space now becomes a Viewpoint from which to consider

Max's next move. Intuitively, when we say In 1991, he would move ..., we

are presenting 1991 as a future with respect to 1990. The 1990 space ('Max in

Rome') becomes a Viewpoint from which to set up the next Focus (and Event)

space, 1991, with the content 'Max move to Venice'. We could have said the

'same' thing differently by using the Base (present time) as a Viewpoint: In

1991, Max moved to Venice.

5) The last sentence, He would then have lived a year in Rome, keeps

1990 as the Viewpoint, and 1991 as the Focus, while using an Event space

('live a year in Rome') which is past time relative to the Focus 1991.

Schematically, the space configuration develops as follows with

successive shifts of Event, Focus, and Viewpoint:

7Base, Vpt, Focus, Event

Base, Vpt, Focus

Event

'Max 23'

'live

abroad'

⇒

Base, Vpt

Focus, Event

1990, live in Rome

⇒

⇒

8Base

Vpt

1990, live in Rome

1991, move Venice

Focus, Event

Vpt

1990, live in Rome

1991, move Venice

Focus

Event

live a year

in Rome

Base

⇒

The virtue of this type of cognitive organization is to allow local

manipulation of the spaces without losing sight of the entire configuration.

Since time is the relevant dimension here, we need some indication of the time

relationship between spaces. Typically, tense will provide us with indications of

9

relative time relationship. Cutrer (1994) proposes putatively universal semantic

tense-aspect categories, with language specific means of expressing some of

their combinations. She also introduces a crucial distinction: new structure

introduced into spaces may be marked as FACT or as PREDICTION,

depending on the semantic tense-aspect. Much of Cutrer's work is devoted to

establishing the constraints on the space configurations that are set up in this

way. The (putatively universal) categories constrain the configuration in

specific ways. For instance, in the case of PAST, we have:

PAST applied to space N indicates that:

i) N is in FOCUS

ii) N's parent is VIEWPOINT

iii) N's time is prior to VIEWPOINT (i.e. prior to N's parent)

iv) events or properties represented in N are FACT (in relation to the

parent VIEWPOINT space)

These general constraints are coded grammatically by languages in

different ways. So what we call the grammatical 'simple past', 'past participle',

and so on, are distinguished from the semantic PAST, which specifies mental

space relationships. English has the following coding system:

PAST is coded by the simple past (lived, went, brought), or by have

+ past participle if the verb is in infinitival position ('will have forgotten', 'may

have left', 'claims to have forgotten'). Code: Verb+past or have + (Verb+

past participle)

FUTURE is coded by will + Verb.

0

The construction of connected spaces, with viewpoint and focus shifts

is reflected in the language code by retracing the path from the Base to the

Focus space, using grammatical tenses.

In our example, when the sentence In 1991, he would move to Venice

comes into the discourse, K is the FOCUS/EVENT space, N (1990) is the

VIEWPOINT space, and M is the BASE. The grammatical coding reflects the

path followed from the BASE to the FOCUS:

Base Space M ---PAST---> Viewpoint Space N ---FUTURE----> Focus Space K

The coding will appear on the verb move , because that verb is

introducing new structure into the current Focus space. The FUTURE

connection of K to N will be coded in English by [will + Verb move ]. The

PAST connection of N to M will be coded by the simple past. The full coding

from Base to Focus is compositional:

simple past + [ will + Verb move ]

⇒ (past + will) + move

⇒ would move

Languages have different ways of coding the Time Path, and grammar

may highlight some aspects of the path, while underspecifying others. What

seems to be universally available is the construction of paths, and the shifts of

Focus and Viewpoint within the dynamic evolving mental space configuration.

General principles govern the ways in which Focus and Viewpoint (and

even Base) are allowed to shift. Cutrer (1994) proposes detailed principles of

discourse organization, which include constraints like the following:

1

- only one FOCUS, one BASE at any given moment of the discourse

interpretation;

- new spaces are built from BASE or FOCUS;

- FOCUS can shift to EVENT, BASE, or previous FOCUS;

- VIEWPOINT can shift to FOCUS or BASE.

The account of tense developed by Dinsmore and Cutrer explains why

tense does not directly reflect conceptual time as one might think (and as many

semantic accounts suggest). Instead, the grammar of tense specifies partial

constraints on time and fact/prediction status that hold locally between mental

spaces within a discourse configuration. We may obtain actual information

about time by combining this with other available pragmatic information.

Accordingly, the same tense may end up indicating very different objective time

relations relative to the speech event:

The boat leaves next week.When he comes tomorrow, I'll tell him about the party.If I see him next week, I'll ask him to call you.

[the "present" tense in the above corresponds to a "future" time]

I'm walking down the street one day when suddenly this guy walks upto me...

He catches the ball. He runs. He makes a touchdown. (morming-aftersports report)

[the "present" tense here corresponds to a "past" event]

Do you have a minute? I wanted to ask you a question.I wish I lived closer to my family, now.If I had time now, I would help you.

[the "past" tense corresponds to a "present" time]

If I had the time next week, I would go to your party.I can't go to the concert tonight. You'll have to tell me how it was.

["past" tense corresponds to a "future" time]

2

That will be all for now.He's not on the train. He will have missed it.

[ "future" tense corresponds to a "present" time]

More generally, tenses are used not just to reflect local time relations

between neighboring spaces, but also to reflect epistemic distance, i.e. whether a

space is hypothetical or counterfactual with respect to its parent space. The coding

system remains the same, and a particular tense sequence may reflect both time and

epistemic distance. Here are some examples offered by Sweetser (1996):

If you have Triple-A, then if you go to a telephone, you can solve your problem.

If you had Triple-A, then if you went to a telephone, you could solve yourproblem.

If you had had Triple-A, then if you'd gone to a telephone, you could have solvedyour problem.

We can interpret all three as referring to present time, but with different epistemic

stances. The first is neutral as to the chances that you have Triple-A. The second

suggests that maybe you don't have it. And the third is counterfactual - "you don't

have Triple-A, but if you did ...". Alternatively, one could interpret the second

sentence as referring to a past event and being neutral as to what happened, and as

to whether you had Triple-A, and the third sentence as referring to a past event, and

being counterfactual. The embedded tenses (go, went, had gone and can solve,

could solve, could have solved ) reflect the full epistemic and time path from the

Base, regardless of the corresponding objective time.

Mood (subjunctive vs. indicative) can serve to indicate distinctions in space

accessibility. So, for example, a sentence like Diogenes is looking for a man who

is honest opens a space in which 'Diogenes finds an honest man'. Because of the

Access principle, that was discussed earlier, the description a man who is honest

can either access a new element directly in that space, or can identify a new element

3

in the Base, and access its counterpart in the 'look for' space. The first accessing

path corresponds to a non-specific interpretation: any honest man will do. The

second accessing path corresponds to a specific reading: there is a particular honest

man that Diogenes is looking for. In French, the equivalent of the verb copula is

can be marked as either indicative or subjunctive:

Diogène cherche un homme qui est honnête. [Indicative]

Diogène cherche un homme qui soit honnête. [Subjunctive]

The first sentence with the indicative allows both accessing paths, as in English,

with perhaps a preference for access from the Base (the specific interpretation).

The second sentence on the other hand allows only direct access to an element in

the 'look for' space, i.e. the non-specific reading. This is because the subjunctive

forces the description to be satisfied in the embedded 'look for' space.

A range of intricate space accessibility phenomena linked to grammatical

mood is studied in Mejías-Bikandi (1994, 1996). Rich aspectual phenomena,

involving spaces and viewpoint are discussed in Doiz-Bienzobas (1995).

MAPPINGS AND CONCEPTUAL INTEGRATION

Language, on the surface, seems to have its own very special principles,

structures, and formal constraints. It has been studied extensively as an

autonomous product of the human mind, and claims have even been made that this

autonomy is reflected at the biological level in the form of specialized, innately

based, genetically transmitted, neurobiological structures. In contrast, cognitive

linguistics has repeatedly uncovered, behind the idiosyncrasies of language,

evidence for the operation of more general cognitive processes. Mappings between

4

mental spaces are part of this general organization of thought. Although language

provides considerable data for studying such mappings, they are not in themselves

specifically linguistic. They show up generally in conceptualization, and there is no

reason to think that they are limited to humans. A striking case of a general

cognitive operation on mental spaces, that is reflected in many language

phenomena, but not restricted to such phenomena, is conceptual integration. There

has been a good deal of recent research on conceptual integration and blending.5 In

this section, I give a quick overview of some of the results Mark Turner and I have

obtained in our joint work on this topic.

Conceptual integration consists in setting up networks of mental spaces

which map onto each other and blend into new spaces in various ways. In

everyday thinking and talking, we use conceptual integration networks

systematically in the on-line construction of meaning. Some of the integrations are

novel, others are more entrenched, and we rarely pay conscious attention to the

process, because it is so pervasive.

A basic conceptual integration network contains four mental spaces. Two

of these are called the input spaces, and a cross-space mapping is established

between them. The cross-space mapping creates, or reflects, more schematic

structure common to the inputs. This structure is constructed in a third space,

called the generic. A fourth space, called the blend, arises by selective projection

from the inputs. It develops emergent structure in various ways, and can project

structure back to the rest of the network.

5 Fauconnier and Turner ([in press], 1994, 1996, in preparation), Coulson (1995,1997), Oakley(1995), Robert (to appear), Mandelblit (1997), Mandelblit and Zachar (in press), Turner (1996).

5

There are simple everyday non-linguistic examples of this around us all the

time. For instance, we can create new activities by blending known activities with

new environments. Coulson (1997) considers the case of children in a dormitory

inventing a game, based on basketball, in which you must throw a crumpled up

sheet of paper into a wastepaper basket. This new game is a blend. One input is

partial knowledge of basketball, the other input is the trash disposal situation with

crumpled paper, a wastepaper basket, and so on. The partial mental mapping

relates a ball to crumpled paper, a basketball basket to a wastepaper basket, players

to children. In the new game, as defined by the blend, some properties are

projected from the 'basketball' input (scoring points when the ball falls into the

basket, opponents in a game, winning and losing, ...), some properties are

projected from the 'trash disposal' input (the basket is on the floor, not high up in

the air, the ball has specific properties of crumpled paper, etc.), and some

properties are shared by the two inputs (throwing a projectile into a receptacle).

Many other properties of the game will emerge from affordances of the context

(particular ways of throwing, conventions for scoring, fouls, etc.). The generic

space linked to the cross-space mapping in this case is the more schematic situation

of throwing some object into a container. This very simple example illustrates

central properties of integration, in particular the fact that it is creative ( a new

activity, different from basketball, and different from throwing away paper, is

produced) and underspecified (there is more than one way to project from the

inputs, and more than one possible emergent structure).

6

The overall result of this dynamic process is a network of the following

form:

Input I Input I 1

2

Blend

Generic Space

• Cross-space mapping

• Selective projection from Inpu

• Composition, Completion, Elab

• Emergent structure

• Integration

In the example, input I1 is partial structure from basketball, input I2 is partial

structure from trash disposal, the generic space is the highly schematic throwing of

an object into a container, and the blend is the emergent game of trashcan

basketball.

Cognitive work on such a network consists in aligning the input spaces and

developing a corresponding generic space, projecting selectively into the novel

blended space, and structuring the blended space through pattern completion and

elaboration (by mental simulation or actual action). Integration of this type occurs

in many cases of action and design, such as computer interfaces or automatic bank

tellers. The Macintosh desktop interface integrates two previously known inputs

(computer commands, and office work), by mapping them onto each other

metaphorically, and integrating them into a novel integrated conceptual and physical

space (the interface, with its specific properties). Fauconnier and Turner (in press)

show in some detail how blending operates here, and how it conforms to a general

7

set of optimality principles. In the same vein, Barbara Holder (1997) discusses the

nice example of ATM's, where a cross-space mapping connects the inputs of

computer manipulation and banking activity, and the ATM itself integrates aspects

of both inputs in a physically and conceptually novel design.

Many linguistic constructions originate in conceptual blends, and reflect

them formally in systematic ways. The most transparent case is perhaps noun

compounding, where an integration is reflected by a compound consisting of two

nouns linked respectively to an element in each of the inputs. For example,

Coulson's game with the crumpled paper in the dorm could be called trashcan

basketball. One noun is the name of the game in one input, the other noun is linked

to a salient element in the other input. By picking other elements, the same game

might get called, more or less felicitously, dorm basketball, basketball paper-

throwing, wastepaper ball, and so on. Almost all cases of noun-noun or adjective-

noun compounding involve some degreee of conceptual integration. It is only in

the simplest possible default cases (and perhaps not even in those cases) that such

compounds reduce to boolean union of properties. So, for example (Turner and

Fauconnier (1995)), if we call a big american car a land yacht, we will be mapping

two inputs (travel on land, travel on sea), with counterparts such as 'vehicle/boat',

'driver/skipper', 'road/ocean', and selectively projecting into a blend, where some

features of yachts now apply to certain automobiles:

8

land

owner

skipper

roadcourse

dr iver

yacht

tycoon

water

car

rich

person

highway

We can see the Access principle at work here, since the term 'yacht' is being

used to access its counterparts in the other spaces. We also see that the compound

is formed by using names of elements which are not counterparts in the cross-space

mapping (land and yacht).

Adjectives typically trigger integration processes. When we are concerned

with a child playing at the beach with a shovel, and say things like The child is

safe, The beach is safe, The shovel is safe, there is no fixed property that "safe"

assigns to child, beach, and shovel. The first means that the child will not be

harmed, but so do the second and the third - they do not mean that the beach will

not be harmed or that the shovel will not be harmed. Safe does not assign a

property; it prompts us to evoke scenarios of danger appropriate for the noun and

the context. We worry about whether the child will be harmed by being on the

beach or by using the shovel. Technically, the word "safe" evokes an abstract

frame of danger with roles like victim, location, and instrument. Modifying the

noun with the adjective prompts us to integrate that abstract frame of danger and the

9

specific situation of the child on the beach into a counterfactual event of harm to the

child. We build a specific counterfactual scenario of harm in which child, beach,

and shovel are assigned to roles in the danger frame. Instead of assigning a simple

property, the adjective is prompting us to set up a conceptual integration network

where the inputs are, on the one hand, a frame of danger, and on the other, the

specific situation of the child on the beach with a shovel. The output of the

integration (the blend) is the counterfactual scenario in which the child is harmed.

The word "safe" implies a disanalogy between the counterfactual blend and the

specific input, with respect to the entity designated by the noun. If the shovel is

safe, it is because in the counterfactual blend it is too sharp, but in the specific

situation it is too dull to cut.

This process is general, and underspecified, as we see by assigning the

same values to different frame roles in the relevant scenario. In The shovel is safe,

the child is the victim in the blend if we are concerned about the shovel's injuring

the child, but the shovel is the victim in the blend if we are concerned about the

child's breaking the shovel. Furthermore, any number of roles can be recruited for

the danger frame. In the counterfactual blend for "the jewels are safe," the jewels

are neither victim nor instrument; they have the role possession and the owner is the

victim. If we ship the jewels in packaging, then the counterfactual blend for "the

packaging is safe" has jewels as victim, external forces as cause of harm, and

packaging as barrier to external forces. Other examples showing the variety of

possible roles would be "Drive at a safe speed," "Have a safe trip," and "This is a

safe bet."

Even more elaborate blends, involving several roles, are constructed for

other syntactically simple expressions, like "The beach is shark-safe" versus "The

beach is child-safe." In the context of buying fish at a supermarket, the label on the

0

tuna can read, "This tuna is dolphin-safe" to mean that the tuna was caught using

methods that prevent accidents from happening to the dolphins. This blend looks

more spectacular, but is constructed using the same integration principles as in the

unremarkable "safe beach" or "safe trip."

"Safe" is not an exceptional adjective, with special semantic properties. The

principles of integration are needed quite generally. Color adjectives, for instance,

require non-compositional conceptual integration. "Red pencil" can be taken to

mean a pencil whose wood has been painted red on the outside, a pencil that leaves

a red mark (the lead is red, or the chemical in the pencil reacts with the paper to

produce red, or . . .), a pencil used to record the activities of the team dressed in

red, a pencil smeared with lipstick, not to mention pencils used only for recording

deficits. Theories of semantics typically prefer to work with examples like "black

bird" or "brown cow" since these examples are supposed to be the prototypes of

compositionality of meaning, but in fact even these examples illustrate complicated

processes of conceptual integration. From a psychological perspective, the key

point in all this is the uniformity and avalailability of the cognitive operation

(conceptual integration) in the full array of cases - those that look unexceptional and

those that are more evidently non-compositional, such as:

"The political war over the future of Medicare, a cannonade oforatory and statistical rotten tomatoes that has already thoroughlyspattered both parties, is about to resume."6

The example of safe shows how an apparently simple expression can

covertly introduce counterfactual scenarios and mappings. In fact, counterfactuals,

6 Michael Wines, "Political Stakes Increase In FightTo Save Medicare," New York Times, Monday, 3 June 1996.

1

which play a key role in our reasoning capacities, are themselves very good

examples of conceptual integration. The meaning of expressions like If I were you,

I would hire me involve a mapping from a reality-space, where 'you' is doing the

hiring, to a counterfactual space in which the speaker's dispositions, but not her/his

situation have been transferred to the addressee you . The space-connection

beween speaker in reality, and the addressee in the counterfactual, allows the

counterfactual employer to be identified as "I" in virtue of the Access principle.

Such counterfactuals are also clearly analogical. They invite an analogical

interpretation in an intuitively obvious way: compare the present situation in which

you are doing the hiring to one in which I am in your place, confronted with the

same decisions to make, and the same candidates. Then export some features (like

'who I would hire') from one situation to the other. However, what we find in the

linguistic expression, "I would hire me" , cannot be a reflection of either one of

these situations. It is not being suggested that the employer should hire herself, or

that the employee should take power and hire himself. The linguistic expression

signals a Blend, which has been set up by the cross-space analogical mapping

between the two situations. In the Blend we find partial projection of Input 1, the

employer and the employee (me) being hired, and partial projection of Input 2, the

speaker's dispositions and decisions. The projections, underspecifications, and

optimality constraints that govern them, are studied in some detail in Fauconnier

(1997), Turner and Fauconnier (in press). It turns out that all counterfactuals are

analogical, and require some degree of blending. Just like the cases of the

compounds, some cases of counterfactuals are more visible than others. Eve

Sweetser came across the following comment on Dostoievski:

"If Dostoevsky had lived in America, and had had a sunnierdisposition, he might have been Emerson."

And the following appeared in the Los Angeles Times:

2

[A woman who had already been in a coma for ten years was raped by ahospital employee, and gave birth to a child. A debate ensued concerning whetherthe pregnancy should have been terminated. The article reporting the case endedwith a comment by law professor Goldberg.]

—"Even if everyone agrees she [the comatose woman] was pro-life at 19,she is now 29 and has lived in PVS [persistent vegetative state] for 10 years. Dowe ask: 'Was she pro-life?' Or do we ask more appropriately: 'Would she be pro-life as a rape victim in a persistent vegetative state at 29 years of life?' "

The inputs are respectively the one in which the young woman falls into a

coma at 19, is raped and becomes pregnant at 29, and the one in which she leads a

normal life, her opinions may change, and she is now 29 years old. This second

input space is of course fictive (counterfactual). Selective projection operates from

these two inputs into the blend. From the first, we project recent events - the rape,

the pregnancy, and also the coma without its consequences. From the second, we

project the typical evolution of a young woman in command of her faculties,

capable of understanding the situation, including the aspects of the situation which

pertain to a person in a coma. In the blend, the young woman is a rape victim, is in

a coma, and has all the faculties and judgment of an ordinary person - namely

herself if things had gone otherwise. Clearly, the blend is not constructing a

'possible world'. Its function is to allow the operation of certain analogies and

lines of reasoning, because it remains linked within the integration network to the

other spaces and to the applicable cultural models.

Metaphor is another area where integration networks are routinely

constructed. Take the stock example This surgeon is a butcher. The statement

underscores the clumsiness of the surgeon, and its undesirable effects. But such

inferences are not simply transferred from the domain of butchers to the domain of

surgery. Butchers are in fact typically quite deft in their own domain of meat-

cutting, and their actions in that domain (producing roasts, steaks, and so on) are

3

considered desirable. In an integration network, two input spaces with very partial

structures from meat-carving and surgery are mapped onto each other, on the basis

of shared generic properties (cutting flesh, sharp instruments, white coat,

professional activity, ...). But neither the clumsiness, nor its catastrophic

consequences, appear in those input spaces. They emerge in the blend. In the

blend, there is projection on one hand of the operating room, the patient, and the

surgeon, and on the other of the butcher's tools, the butcher's methods and manner

of carving, etc. Emergent structure ensues from simulation of this unusual

situation, and we are able to grasp instantly the nefarious effects of the procedure.

The resulting failings of the surgeon, represented with considerable hyperbole in

the blend, are projected back to the input space of surgery, where they yield an

inference of gross incompetence.

Mark Turner (personal communication) offers the following simple example

to illustrate this very general process. The reference is to the stock market, a bull

market until recently. Investors in such a market are commonly called bulls. When

the market showed signs of weakness, a financier, Arnie Owen, was quoted in the

Wall Street journal as saying "Everybody has their horns pulled in". In the input

space of cattle behavior, bulls can't pull in their horns. In the input space of

finance, investors don't have horns, but they can pull back on their investing. In

the blend, the investors end up as bulls with retractable horns. This emergent and

inferentially efficient structure in the blend is not available through direct

source/target metaphorical mapping.

In Fauconnier and Turner (in press, in preparation), we show how

integration allows basic metaphors and metonymies to operate very generally in

conceptual integration networks to produce complex, and often multiple, blends.

This is the case for instance in the well known ANGER as HEAT and FAILING as

4

DYING metaphors. Interestingly, blending allows mapping between inputs that

may clash in certain relevant respects such as intentional, causal, or temporal

structure.

When we examine the full range of conceptual integration networks, we

find that there is no sharp distinction of kind between simple framing, non-typical

framing, analogy, source-target metaphor, and multiple metaphor. Rather, we have

a continuum within the range of networks, with simple Fregean semantics at one

end, and elaborate metaphor at the other, with the same integration principles in all

cases applying to a variety of topological configurations. On this continuum, we

find points that are prototypical in certain respects, and it is those protoypical points

that we associate with the notions, familiar and superficially different in appearance,

of predicate-calculus, framing, analogy, simple metaphor, abstract metaphor. We

have been able to show (Fauconnier and Turner (in preparation)) that the mapping

schemes in all these cases were the same, and that they were compositional in an

interesting cognitive, but not truth-conditional, sense. One may get an intutive

sense of this continuum by looking at a set of examples like the following, ordered

along a dimension of complexity of integration networks: Paul is the father of Sally.

He is my genetic father but not my real father. Zeus is the father of Athena. I am

your father for today [friend of the family taking care of a child for a short time].

George Washington is the father of our country. Newton is the father of modern

physics. Fear, father of cruelty (Ezra Pound). The child is the father of the man.

(Wordsworth).

5

CONCLUSION

I have tried to give an idea of the elaborate backstage cognition that operates

behind everyday utterances through spoken or signed modalities. This brief and

informal overview does not do justice to other extensive work in this area. In

particular, the construction of richer discourse and narrative has not been described

here. Sanders and Redeker (1996), Cutrer (1994), and Mushin (1998) provide

fascinating case studies and elaborations of the mental space model to deal with

them. Rubba (1996) shows how multiple spaces are constructed on the basis of

sparse linguistic information and elaborate, but covert, cultural and cognitive

models. Fridman-Mintz and Liddell (1998) also find that multiple and successive

connected spaces are constructed in short narratives in ASL. Work by Hutchins

(1995) shows how conceptual blending can use material anchors to produce

situated behavior, such as navigation of ships and planes. Coulson (1997)

develops integration accounts of rhetorical strategies, noun and adjective

compounding, and argumentative counterfactuals. Turner (1996), Oakley (1995),

Freeman (1997) develop elaborate analysis of mental space construction and

integration in literary narratives, poetry, and general rhetoric. Zbikowski (1997)

shows how similar processes operate in music. Lakoff and Nunez (in preparation)

find many cases of conceptual blends in mathematics in addition to the ones pointed

out in Fauconnier and Turner (in press).

Two central points stand out in all of this work, and are particularly relevant

for cognitive psychology:

1) Backstage cognition: language works in concert with "behind the

scenes" understandings and cognitive processes. Language is not a representation

of such processes, or a representation of meaning. Rather it serves as a powerful

and directed, but vastly underspecified, set of prompts for triggering the dynamic

6

processing itself, and the corresponding construction of meaning. Backstage

cognition has not received from linguists the attention it deserves because it is

largely unconscious, immediate, and structurally invisible.

2) An especially important set of such processes (that are especially well-

hidden from conscious view) concerns mental spaces and their associated cognitive

operations.

There is little doubt that mental space constructions permeate much of our

thought, our use of language, and other forms of expression. The study of such

general cognitive operations should be a fertile ground for experimental

psychology. The findings in this area converge nicely with those of Bloom

(1974,1991), Barsalou (1996), Mandler (to appear).

7References

Barsalou, L. 1996. Perceptual Symbol Systems. University ofChicago ms.

Bloom, L. 1974. Talking, Understanding, and Thinking. In R.L.Schiefelbusch & L.L. Lloyd (Eds.), Language Perspectives - Acquisition,Retardation, and Intervention. Baltimore, Maryland: University ParkPress.

Bloom, L. 1991. Representation and Expression. In N. Krasnegor,D. Rumbaugh, R. Schiefelbusch, & M. Studdert-Kennedy (Eds.), Biologicaland behavioral foundations for language development. Hillsdale, N.J.:Erlbaum.

Coulson, Seana. 1995. Analogic and metaphoric mapping in blendedspaces. Center for Research in Language Newsletter volume 9, number 1,pages 2-12.

Coulson, Seana. 1997. Semantic Leaps: Frame-Shifting and ConceptualBlending. UCSD Ph. D. dissertation.

Cutrer, M. 1994. Time and Tense in Narratives and EverydayLanguage. Doctoral dissertation, University of Calfornia at San Diego.

Dinsmore, J. 1991. Partitioned Representations. Dordrecht: Kluwer.

Doiz-Bienzobas, A. 1995. The Preterite and the Imperfect inSpanish: Past Situation vs. Past Viewpoint. Ph. D. dissertation.University of California San Diego.

Fauconnier, G. 1994. Mental Spaces. New York: Cambridge UniversityPress. [Originally published (1985) Cambridge: MIT Press.]

Fauconnier, G. 1997. Mappings in Thought and Language. Cambridge:Cambridge University Press.

Fauconnier, G. & E. Sweetser. 1996. Spaces, Worlds, and Grammar.Chicago: University of Chicago Press.

Fauconnier, G. & Turner, M. 1994. Conceptual projection and middlespaces. UCSD: Department of Cognitive Science Technical Report 9401.[Available from http://cogsci.ucsd.edu and fromhttp://www.wam.umd.edu/~mturn]

Fauconnier, G. & Turner, M. (1996). Blending as a central process ofgrammar. In Conceptual Structure, Discourse, and Language, Ed. Adele

8

Goldberg. Stanford: Center for the Study of Language and Information[distributed by Cambridge University Press].

Fauconnier, G. & Turner, M. (in press). Conceptual IntegrationNetworks. Cognitive Science.

Fauconnier, G. & Turner, M. (in preparation). Conceptual Integration.

Freeman, Margaret. 1997. Grounded spaces: Deictic -self anaphors inthe poetry of Emily Dickinson. Language and Literature, (1997) 6:1, 7-28.

Fridman-Mintz, B. and S. Liddell. 1998. Sequencing Mental Spaces inan ASL Narrative. In Conceptual Structure, Discourse, and Language, II, Ed.J.-P. Koenig. Stanford: Center for the Study of Language and Information.

Holder, B. 1997. Blending and Your ATM. UCSD ms.

Hutchins, E. 1995. Cognition in the Wild. Cambridge: MIT Press.

Huumo, T. 1996. A Scoping Hierarchy of Locatives. CognitiveLinguistics 7 (3): 265-299.

Lakoff, G. and R. Núñez. (in press) Cognition for a Mind-basedMathematics. In Lyn English, ed., Mathematical Reasoning: Analogies,Metaphors, and Images. Hillsdale, NJ: Lawrence Erlbaum.

Liddell, Scott K. 1995a. Real, Surrogate and Token Space:grammatical consequences in ASL. In K. Emmorey and J. Reilly (eds.),Language, Gesture,and Space. Hillsdale, NJ: Lawrence ErlbaumAssociates, Publishers, 19-41.

Liddell, Scott K. 1995b. Spatial Representations in Discourse. ms.Gallaudet University.

Mandelblit, Nili and Oron Zachar. (in press). The Notion of Unit andits Development in Cognitive Science. Cognitive Science.

Mandelblit, Nili. 1997. Creativity and Schematicity in Grammar andTranslation: The Cognitive Mechanisms of Blending. Doctoral dissertation,UCSD Cognitive Science.

Mandler, J. to appear. Representation. In Cognition, Perception,and Language, eds. D. Kuhn and R. Siegler. Vol.2 of W. Damon, ed.,Handbook of Child Psychology.

Mejias-Bikandi, E. 1993. Syntax, Discourse, and Acts of Mind: AStudy of the Indicative/Subjunctive in Spanish. Doctoral dissertation,UCSD.

9

Mejias-Bikandi, E. 1996. Space Accessibility and Mood in Spanish.In Fauconnier, G. & E. Sweetser, eds. Spaces, Worlds, and Grammar.Chicago: University of Chicago Press.

Mushin, I. 1998. Viewpoint Shifts in Narrative. In ConceptualStructure, Discourse, and Language, II, Ed. J.-P. Koenig. Stanford: Center forthe Study of Language and Information.

Oakley, Todd. 1995. Presence: the conceptual basis of rhetoricaleffect. Doctoral dissertation, University of Maryland.

Poulin, C. 1996. Manipulation of discourse spaces in ASL. InConceptual Structure, Discourse, and Language, ed. Adele Goldberg.Stanford: Center for the Study of Language and Information [ distributedby Cambridge University Press].

Robert, A. (to appear) Blending in Mathematical Proofs. ConceptualDiscourse, Structure, and Language II. Ed. Jean-Pierre Koenig. Stanford:Center for the Study of Language and Information

Rubba, J. 1996. Alternate Grounds in the Interpretation of DeicticExpressions. In Fauconnier, G. & E. Sweetser, eds. Spaces, Worlds, andGrammar. Chicago: University of Chicago Press.

Sanders, J. and G. Redeker. 1996. Perspective and theRepresentation of Speech and Thought in Narrative Discourse. InFauconnier, G. & E. Sweetser, eds. Spaces, Worlds, and Grammar.Chicago: University of Chicago Press.

Shastri, L. and D. Grannes. 1996. A Connectionist Treatment ofNegation and Inconsistency. Proceedings of the 18th Annual Conference ofthe Cognitive Science Society. Lawrence Erlbaum.

Sweetser, E. 1996. Mental Spaces and the grammar of conditionalconstructions. In Fauconnier, G. & E. Sweetser, eds. Spaces, Worlds, andGrammar. Chicago: University of Chicago Press.

Turner, Mark. 1996. The Literary Mind. New York: Oxford UniversityPress.

Turner, Mark and Gilles Fauconnier. 1995. Conceptual Integration andFormal Expression. Journal of Metaphor and Symbolic Activity, volume 10,number 3, pages 183-204.

Turner, Mark and Gilles Fauconnier (in press). Conceptual Integrationin Counterfactuals. In Conceptual Structure, Discourse, and Language, II, Ed.J.-P. Koenig. Stanford: Center for the Study of Language and Information.

0

Van Hoek, K. 1996. Conceptual Locations for Reference inAmerican Sign Language. In Fauconnier, G. & E. Sweetser, eds. Spaces,Worlds, and Grammar. Chicago: University of Chicago Press.

Zbikowski, Lawrence. 1997. "Conceptual blending and song."Manuscript. University of Chicago.